>>>> >>> I'm having issues where hcidump seems to be losing packets. I'm >>>> >>> trying to collect traces of avtest running. Avtest properly prints >>>> >>> out that it sent a packet and received a reply, but I don't always see >>>> >>> it in hcidump. Seems if I run avtest a lot successively, it does show >>>> >>> up. I'm running a 2.6.33.20 kernel with hcidump-2.3. Let me know if >>>> >>> the kernel is just too ancient, but I'd like to know if others have >>>> >>> seen this type of issue. >>>> >> >>>> >> actually hcidump can not loose any packets. Write them into a file in >>>> >> BTSnoop format and double check with Wireshark or the Frontline Viewer. >>>> >> >>>> >> The kernel is always sending the exact copy to the hcidump sockets and >>>> >> thus there is no way you loose packets. The only explanation would be >>>> >> that L2CAP socket has a bug and swallows it. Have you checked that >>>> >> avtest actually reports L2CAP send errors. >>>> > >>>> > I was saving in btsnoop and looking at it in wireshark. And verified >>>> > that the other side got the command by enabling debug mode on >>>> > bluetoothd. But now I see the packet is sort of there, but it's >>>> > malformed! >>>> > >>>> > Bad: >>>> > < 02 01 20 07 00 03 00 40 00 00 06 04 >>>> >> 02 01 20 07 00 03 00 40 00 03 06 31 >>>> > >>>> > Good: >>>> > < 02 01 20 07 00 03 00 42 00 00 06 04 >>>> >> 02 01 20 07 00 03 00 42 00 03 06 31 >>>> > >>>> > Somehow one byte is wrong (40 bad, 42 good). Any ideas on that? Let >>>> > me know if you'd like the logs, but I verified the --raw output of >>>> > hcidump had the same issue. >>>> >>>> BTW, I would say the issue exists more when SDP is going on (as when I >>>> run the test a lot before a disconnect, it starts working -- no more >>>> SDP later on). And we can see that it classified the packet as SDP, >>>> not AVDTP. I'm tracking the hcidump code now, not entirely sure how >>>> data goes into one or the other. >>> >>> the hcidump code is pure raw data. It is the same data that goes to the >>> driver. Trust me here, that is how the kernel works, it clones the skb >>> before sending to the driver and sends a copy back to hcidump. So you >>> will not find a bug in the hcidump raw data. >>> >>> The only way something can go wrong if L2CAP has an issue. Are you using >>> L2CAP basic modem or enhanced retransmission? >> >> Looks like ERTM was added in 2.6.36, and I didn't backport it, so must be basic. > > Unless it was added earlier. I'm really not familiar enough. Just > running avtest.c, sending invalid commands! OK, finally figured it out. Wireshark was at fault [1]. The fix has been in Wireshark trunk for a while, but their latest stable build doesn't have that. Basically, it was picking the last packet for a given CID. If I ran it once, the AVDTP packets went first, followed by SDP so everything was parsed as SDP. [1] https://bugs.wireshark.org/bugzilla/show_bug.cgi?id=6619 -- To unsubscribe from this list: send the line "unsubscribe linux-bluetooth" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
